Funnel metrics: measure the movement, not the moment

Every company we've ever audited has a funnel dashboard. Most of them are museums: a row of conversion percentages, a line trending somewhere, a red number somebody's boss dislikes. The dashboard gets glanced at on Mondays, and the same arguments happen every week — marketing says leads are fine, sales says leads are bad, product says the signup flow is fine actually, and the meeting ends with someone promising to "dig into the data."

The problem is rarely the data. It's that the funnel was designed to display numbers rather than to locate the constraint. A funnel that informs decisions has three properties most dashboards lack: everyone agrees on stage definitions, the metrics measure movement rather than moments, and there's a weekly ritual that forces one constraint to the surface. This is the system we install, and it's the same skeleton whether the funnel ends in a checkout, a booked demo or a shipped SaaS seat.

Stage definitions: a negotiation you have to have once

Before any metric work, write down the funnel stages and get product, marketing and sales to sign the definitions. This sounds bureaucratic; it is actually the highest-leverage conversation in the whole system, because most funnel arguments are definition arguments wearing metric costumes.

The rules for definitions that hold:

Stages are observable events, not moods. "Activated" is an event or a small, explicit set of events (connected a bank, ran a first report), not "the user seems engaged." If activation can't be written as a query, it isn't a stage — it's a vibe, and vibes don't reconcile across teams.
Each stage has one owner. Ownership doesn't mean blame; it means the person who speaks first when that stage's number moves. A stage owned by "growth" is owned by no one.
Entry and exit are both defined. "MQL" fails not because leads are bad but because teams define entry ("downloaded the guide") without exit ("spoke to sales or decayed after 30 days"). Undefined exits are how stages silently accumulate corpses that poison every conversion rate downstream.
Definitions change rarely and loudly. Annotate the dashboard when a definition changes. A metric whose meaning shifted in March is a metric lying about February.

When we rebuilt analytics for the Northwind Ledger dashboard, the single most valuable artifact wasn't the charts — it was a one-page stage-definition document that ended the weekly "are these leads good" argument in week two.

Measure movement: velocity over snapshots

The snapshot conversion rate — "4.2% of visitors start a trial" — is the most quoted and least decision-useful number in growth. It mixes cohorts, lags reality, and hides the dynamics that matter. Movement metrics fix this:

Stage-to-stage conversion, cohorted. Don't ask "what share of signups activate?" Ask "of everyone who signed up in week 12, what share activated within 14 days?" Cohorting removes the denominator pollution that makes snapshot rates lie — mature funnels look artificially good because old cohorts keep converting slowly, and young companies look artificially bad because their cohorts haven't had time.

Time-in-stage (velocity). How long does a lead sit in trial? How many days from signup to activation? Velocity is the most underused metric in the stack, and often the most actionable: a funnel where healthy conversions happen but slowly has a friction problem, not a demand problem — and friction problems are the kind you can fix this quarter. Velocity also predicts: cohorts whose early velocity degrades almost always retain worse, weeks before retention curves make it official.

Volume at entry, quality-adjusted. Entry volume is only meaningful paired with a downstream outcome per entry cohort. A 40% jump in signups that halves activation means the channel changed, not that the funnel grew. We went deep on the accounting discipline behind this in the attribution article; the short version: entry metrics without cohort outcomes are theatre.

Leakage concentration. Not "where do we lose people" as a flat list — every funnel loses people everywhere — but where the loss is asymmetric relative to effort required to fix it. A 5-point drop at a stage with known, fixable causes beats a 20-point drop at a stage requiring a rebuild. The metric's job is triage.

The weekly funnel review: one constraint, one owner, one bet

The instrument that turns metrics into growth is boring: a thirty-minute weekly review with a fixed agenda. The format we leave with clients:

The walk (10 min). Stage by stage, versus trailing 4-week: volume in, conversion to next, velocity. One owner speaks per stage, one minute each. No diagnosis yet — just the movement.
Identify the constraint (10 min). A growth system has exactly one binding constraint at a time — the stage where improvement currently yields the most. Find it. Not the stage that's worst-looking; the stage where the combination of gap and fixability is best. Everything else is explicitly parked.
One bet, one owner, one date (10 min). The week's experiment targets the constraint, with a pre-registered success criterion — the discipline we described in designing CRO experiments you can believe. If the constraint isn't addressable this week (waiting on an integration, say), the bet targets the next-best constraint, and the blocker gets an owner with a date.

Three failure modes to refuse: reviewing more than one funnel per meeting (pick the one that feeds revenue this quarter); letting the meeting generate a backlog instead of a bet (backlogs are where urgency goes to sleep); and skipping when the numbers are boring. Boring weeks are when slow leaks get caught — the churn-cohort creep, the velocity drift — because dramatic weeks are full of explanations and boring weeks are full of truth.

Building the dashboard: less, but load-bearing

The dashboard itself should fit a screen and answer four questions at a glance: how much entered, how far they got, how fast they moved, and whether this week's cohort differs from recent ones. Concretely:

One cohort table (signup week × stages reached) beats any number of charts. Conditional formatting does the analysis; arrows and sparklines do decoration.
Velocity as a distribution, not an average. Median and 90th percentile time-to-activate. Averages in funnels are liars — three fast paths and one dead end average into a shrug.
Annotations, always. Releases, pricing changes, campaign launches, outages. An unannotated dashboard teaches the team to invent stories; an annotated one teaches them to check dates first. This is part of the broader tracking-plan discipline — the plan is what makes the numbers worth arguing about.

Tooling is near-irrelevant at this altitude. A warehouse-backed BI tool is lovely; a disciplined spreadsheet updated weekly is sufficient. I've watched a £40k dashboard lose arguments to a spreadsheet with honest definitions, weekly. The load-bearing parts are the definitions, the cohorting, and the ritual — everything else is furniture.

A closing word on honesty

The deepest discipline in funnel metrics is resisting the local optimum of improving the numbers instead of the funnel. Raising trial starts by loosening what "start" means. Improving activation by excluding the cohort that never would. Every metric in a funnel can be improved by fraud — most of it accidental, self-deceiving fraud — and the only defence is the pre-registered definition sheet and a team culture where changing a definition mid-quarter is a bigger violation than a red number.

Get that right and the funnel becomes what it should be: not a report on the business, but an X-ray of where effort belongs next. That's the core of how our growth practice runs engagements — instrumentation before experimentation — and if your funnel review currently produces arguments instead of bets, that's fixable fast. The engagement models are here; the brief form is ten minutes well spent.

Key takeaways

Agree stage definitions once, across teams: observable events, single owners, defined exits, and annotated changes. Most funnel arguments are definition arguments.
Measure movement, not moments: cohort-based stage conversion, time-in-stage velocity, quality-adjusted entry volume, and leakage triaged by fixability.
Run a thirty-minute weekly review with a fixed agenda that surfaces one binding constraint and ends with one bet, one owner, one date — never a backlog.
Keep the dashboard to a cohort table, velocity distributions and annotations. The ritual and definitions are load-bearing; the tooling is furniture.
Defend the numbers from well-meaning fraud: changing a definition mid-quarter is worse than a red number.

FAQ

How many stages should our funnel have? Fewer than you think — typically four to six from first touch to revenue. Each stage must clear the bar of being an observable event with an owner, and if two stages always convert at roughly the same rate, they're one stage cosplaying as two. Add a stage only when it changes what you'd do; a stage that never gets its own interventions is dashboard decoration.

What if our sales cycle is too long for weekly reviews to matter? Long cycles change the cadence, not the system. Keep the weekly review but weight it toward leading velocity metrics — time-to-first-meeting, stage-dwell time, early-cohort progression — which move weekly even when revenue moves quarterly. The month-end view carries the lagging outcomes. If you abandon the ritual because revenue is slow, you lose the early-warning system exactly where cycles are longest.

Activation means something different for every user segment. Help? Then your product has multiple activation events, and the honest model is a per-segment activation definition with a headline number that's an explicit blend. What doesn't work is a mushy single definition that averages segments into meaninglessness. Define each segment's "first payoff" event, measure each cohort against its own, and let the blended number be a computed convenience — never the other way around.

When is it worth building a proper warehouse setup versus staying in the product analytics tool? When one of three things breaks: you need to join product events to revenue or CRM data the tool can't see; cohort definitions disagree across tools and eat meeting time; or query limits force you to sample. Until then, the ceiling on your funnel work is definitions and ritual, not infrastructure. Buy the warehouse to solve a named problem, not to feel serious.